This paper introduces posterior mean matching (PMM), a new method for generative modeling that is grounded in Bayesian inference. PMM uses conjugate pairs of distributions to model complex data of various modalities like images and text, offering a flexible alternative to existing methods like diffusion models. PMM models iteratively refine noisy approximations of the target distribution using updates from online Bayesian inference. PMM is flexible because its mechanics are based on general Bayesian models. We demonstrate this flexibility by developing specialized examples: a generative PMM model of real-valued data using the Normal-Normal model, a generative PMM model of count data using a Gamma-Poisson model, and a generative PMM model of discrete data using a Dirichlet-Categorical model. For the Normal-Normal PMM model, we establish a direct connection to diffusion models by showing that its continuous-time formulation converges to a stochastic differential equation (SDE). Additionally, for the Gamma-Poisson PMM, we derive a novel SDE driven by a Cox process, which is a significant departure from traditional Brownian motion-based generative models. PMMs achieve performance that is competitive with generative models for language modeling and image generation.

Deep neural networks (DNNs) are easily fooled by adversarial perturbations that are impercep-tible to humans. Adversarial training, a process where adversarial examples are added to the training set, is the current state-of-the-art defense against adversarial attacks, but it lowers the model's accuracy on clean inputs, is computationally expensive, and offers less robustness to natural noise. In contrast, energy-based models (EBMs), which were designed for efficient im-plementation in neuromorphic hardware and physical systems, incorporate feedback connections from each layer to the previous layer, yielding a recurrent, deep-attractor architecture which we hypothesize should make them naturally robust. Our work is the first to explore the robustness of EBMs to both natural cor-ruptions and adversarial attacks, which we do using the CIFAR-IO and CIFAR-IOO datasets. We demonstrate that EBMs are more robust than transformers and display comparable robustness to adversarially-trained DNNs on gradient-based (white-box) attacks, query-based (black-box) attacks, and natural perturbations without sacrificing clean accuracy, and without the need for adversarial training or additional training techniques.

Machine learning classifiers have been demonstrated, both empirically and theoretically, to be robust to label noise under certain conditions --- notably the typical assumption is that label noise is independent of the features given the class label. We provide a theoretical framework that generalizes beyond this typical assumption by modeling label noise as a distribution over feature space. We show that both the scale and the \emph{shape} of the noise distribution influence the posterior likelihood; and the shape of the noise distribution has a stronger impact on classification performance if the noise is concentrated in feature space where the decision boundary can be moved. For the special case of uniform label noise (independent of features and the class label), we show that the Bayes optimal classifier for $c$ classes is robust to label noise until the ratio of noisy samples goes above $\frac{c-1}{c}$ (e.g. 90\% for 10 classes), which we call the \emph{tipping point}. However, for the special case of class-dependent label noise (independent of features given the class label), the tipping point can be as low as 50\%. Most importantly, we show that when the noise distribution targets decision boundaries (label noise is directly dependent on feature space), classification robustness can drop off even at a small scale of noise. Even when evaluating recent label-noise mitigation methods we see reduced accuracy when label noise is dependent on features. These findings explain why machine learning often handles label noise well if the noise distribution is uniform in feature-space; yet it also points to the difficulty of overcoming label noise when it is concentrated in a region of feature space where a decision boundary can move.

We tackle the problem of analyzing and retrieving technical drawings. First, we introduce DeepPatent, a new large-scale dataset for recognition and retrieval of design patent drawings. The dataset provides more than 350,000 design patent drawings for the purpose of image retrieval. Unlike existing datasets, DeepPatent provides fine-grained image retrieval associations within the collection of drawings and does not rely on cross-domain associations for supervision. We develop a baseline deep learning model, named Patent-Net, based on best practices for training retrieval models for static images. We demonstrate the superior performance of PatentNet when trained on our fine-grained associations of DeepPatent against other deep learning approaches and classic computer vision descriptors. With the introduction of this new dataset, and benchmark algorithms, we demonstrate that the analysis and retrieval of technical drawings remains an open challenge in computer vision; and that patent drawing retrieval provides a real-world testbench to spur research.

The convenience of cell-phone cameras has made them one of the most common ways by which people document their lives, whether it is everyday pleasures or celebrations. With thousands of images, it might prove to be a daunting task to organize them by hand. When applying automated algorithms to help us, we would like to have both images that are dear to us but are also of good quality. In this paper we explore the performance of the MobileNet CNN architecture, and the different design (inputs size, and layer depth) choices, in their ability in solving various aesthetic inference task: binary classification, regression, image cropping. We show that the baseline MobileNet architecture achieves near state-of-the-art results for binary classification on the AVA dataset while being more than 10 times smaller and compute efficient. We further show that these models, when trained for fine-grained aesthetics inference, achieve better cropping performance than other aestheticsbased croppers.

The ability to rank images based on their appearance finds many real-world applications such as image retrieval or image album creation. Despite the recent dominance of deep learning methods in computer vision which often result in superior performance, they are not always the methods of choice because they lack interpretability. In this work, we investigate the possibility of improving image aesthetic inference of convolutional neural networks with hand-designed features that rely on domain expertise in various fields. We perform a comparison of hand-crafted feature sets in their ability to predict fine-grained aesthetics scores on two image aesthetics datasets. We observe that even feature sets published earlier are able to compete with more recently published algorithms and, by combining the algorithms together, one can obtain a significant improvement in predicting image aesthetics. By using a tree-based learner, we perform feature elimination to understand the best performing features overall and across different image categories. Only roughly 15 % and 8 % of the features are needed to achieve full performance in predicting a fine-grained aesthetic score and binary classification respectively. By combining hand-crafted features with meta-features that predict the quality of an image based on CNN features, the model performs better than a baseline VGG16 model. One can, however, achieve more significant improvement in both aesthetics score prediction and binary classification by fusing the hand-crafted features and the penultimate layer activations. Our experiments indicate an improvement up to 2.2 % achieving current state-of-the-art binary classification accuracy on the AVA dataset when the hand-designed features are fused with activation from VGG16 and ResNet50 networks.